Avoid stale approved follow-up issues

[wwind123]

Summary

• Process approved-review future follow-ups only from the final approval round
• Stop carrying pre-fix future follow-ups across same-PR fix rounds
• Update tests and docs for final-round-only follow-up handling

Fixes #73

Tests

• /tmp/coding-review-agent-loop/scratch/issue73-venv/bin/python -m pytest tests/test_agent_loop.py

[wwind123]

Implemented issue #73 and opened PR #75.

Changed approved follow-up handling so stale future follow-ups from earlier same-PR fix rounds are not carried forward; only future follow-ups restated in the final approval round are summarized or turned into issues. Updated tests and docs accordingly.

Tests:
/tmp/coding-review-agent-loop/scratch/issue73-venv/bin/python -m pytest tests/test_agent_loop.py - 130 passed

-- OpenAI Codex

[wwind123]

Review: Avoid stale approved follow-up issues (#75)

The fix is minimal and correct. Removing pending_future_followups entirely and switching to approved_followups = round_future_followups is exactly the right approach for issue #73.

Logic review:

The old accumulator pending_future_followups.extend(round_future_followups) in the same-PR fix path meant that when round 1 approved-with-same-PR-fixes also included future follow-ups, those future items were merged with round 2's future follow-ups on final approval. Since round 2 reviewers see the updated code and can restate whatever is still relevant, carrying round 1 items forward created stale noise. The fix correctly trusts the final round's reviewers to surface anything worth tracking.

The discard-on-blocking path (the else branch) already had the right behavior and didn't need a logic change — just a comment cleanup, which reads more accurately now.

Test review:

The three test changes work together:

• test_pr_loop_fix_and_summarize_sends_same_pr_followups_to_coder_then_rereviews: comment count drops from 4 → 3 because round 2 has no future follow-ups, so no summary comment is posted. Correct.
• Renamed test_pr_loop_fix_and_issue_ignores_future_followups_from_pre_fix_round: same setup, final round has no future follow-ups → runner.issues == []. Tests the discarded case cleanly.
• New test_pr_loop_fix_and_issue_uses_only_final_round_future_followups: round 1 has "Stale item…", round 2 has "Add a separate migration dry-run command." — verifies exactly one issue is created for the final-round item and the stale item is absent. This is the core coverage for the new behavior.
• test_pr_loop_fix_and_summarize_uses_only_final_round_future_followups: multi-reviewer case — only "final follow-up" items appear in the summary, "larger follow-up" items from round 1 do not. Correct.

Docs review:

Both README.md and docs/local_agent_loop.md now say "only future follow-ups restated in the final approval round are summarized or turned into issues," which accurately describes the new behavior.

No blocking issues.

-- Anthropic Claude

[wwind123]

Review: Avoid stale approved follow-up issues (#75)

The changes correctly address issue #73 by ensuring that only future follow-ups from the final approval round are processed. This prevents "stale" or redundant suggestions from being carried forward when multiple review rounds occur within the same PR.

Logic Review

• The removal of the pending_future_followups accumulator in run_pr_loop is correct.
• By resetting round_future_followups in each round and only using it when the PR is finally approved, the orchestrator now correctly trusts the final state of the review.
• Discarding items when there are blocking reviews or same-PR fixes is the right behavior, as it allows reviewers to re-evaluate what is still a "future" item once the code has been updated.

Test Coverage

• The updated tests (test_pr_loop_fix_and_summarize_sends_same_pr_followups_to_coder_then_rereview, etc.) correctly verify the expected behavior.
• The new tests (test_pr_loop_fix_and_issue_uses_only_final_round_future_followups and test_pr_loop_fix_and_summarize_uses_only_final_round_future_followups) provide solid coverage for the fix, confirming that stale items are indeed excluded.

Documentation

• Both README.md and docs/local_agent_loop.md have been updated to accurately describe the new behavior.

The implementation is surgical, well-tested, and maintainable.

-- Google Gemini